This is a talk given at the Redwood Center for Theoretical Neuroscience, UC Berkeley on September 5, 2006. Speaker is Peter Latham of
Gatsby Computational Neuroscience Unit, University College London.
A major open question in neuroscience is: "what's the neural code?". The standard approach to answering this, pioneered by Richmond and Optican almost two decades ago , is to record spike trains and compute information under different coding models. Unfortunately, this approach requires huge amounts of data and thus, despite considerable efforts and some success, it is still not clear to what extent the neural code, in mammalian cortex, relies on precise spike timing, and in particular on spike patterns.
An alternative approach follows from the observation that if spike patterns are to carry information, they must be precisely repeatable. We may thus ask the question: does the massively recurrent connectivity that is a salient feature of cortical networks place intrinsic limits on precise repeatability, and thus on the extent to which spike patterns can carry information?
We show here that one can answer this question by measuring the mean increase in the firing rate of an average neuron in response to a single synaptic input. If the mean increase is sufficiently large, then the network must be chaotic at the microscopic level, which in turn precludes precisely repeatable spike trains. To address this experimentally, we use in-vivo patch-clamp recordings from cortical pyramidal neurons in barrel cortex of anesthetized rats, and we find that these networks are highly chaotic. We then perform single neuron simulations with models of biophysically realistic neurons and large network simulations with conductance-based neurons. Consistent with our experiments and with previous studies [2,3], we find the same result: highly chaotic dynamics. Finally, we connect quantitatively the mean increase in firing rate with a lower bound on the precision at which spike timing can carry information.
1. B.J. Richmond and L.M. Optican, J. Neurophysiol. 57:132-46;57:147-61 (1987). 2. C. van Vreeswijk and H. Sompolinsky, Neural Comput. 10:1321-1371 (1998). 3. A. Banerjee, Neural Comput. 13:161-193; 13:195-225 (2001); J. Comput. Neurosci. 20:321-48 (2006).
Contact InformationKilian Koepsell Redwood Center for Theoretical Neuroscience University of California Helen Wills Neuroscience Institute 132 Barker, MC #3190 Berkeley, CA 94720-3190